Anthropic and Trump Administration Near Deal to Lift Fable 5 and Mythos 5 Export Controls

Anthropic PBC and the Trump administration are moving closer to an agreement that would lift restrictions on the company's top two AI models, Fable 5 and Mythos 5, following weeks of talks over security concerns, according to people familiar with the matter.

According to Bloomberg News, which obtained a copy of Lutnick's original letter and spoke with people familiar with the negotiations, Commerce Secretary Howard Lutnick is making progress on addressing the security concerns that triggered the restrictions in the first place. Those people said the controls on both Fable 5 and Mythos 5 could be fully lifted once officials across the administration give their sign-off.

Who Was in the Room

Anthropic co-founder Tom Brown met with Lutnick and other senior administration officials within the past several days, according to Bloomberg's sources, who spoke on condition of anonymity. The company itself did not respond to a request for comment.

A U.S. official told Bloomberg that Commerce is working toward a resolution and is specifically focused on ensuring the entire administration is aligned before any final decision is made. The alignment requirement is currently the critical factor. The White House and other agencies have taken an active role in shaping AI policy, and any deal needs broader sign-off beyond Lutnick's office alone.

What Triggered This

The original restriction, issued roughly two weeks ago in Lutnick's letter to Anthropic, barred foreign nationals from accessing Fable 5 and Mythos 5 over concerns that security guardrails on both systems could be circumvented. Anthropic responded by disabling all global access to the two models, not just foreign access, which shut down paying customers worldwide and accelerated the talks.

Bloomberg described the letter as the most significant U.S. government intervention into an AI company's operations to date. The timing was notable: it came weeks after Anthropic filed confidentially for an IPO, with its latest valuation reported above $900 billion.

Anthropic had been candid about Mythos 5's risks before any government action. The company first shared the model in April with a limited group, explicitly warning that its ability to identify cybersecurity vulnerabilities made broader distribution too risky. Fable 5, released earlier this month as the first public-facing version of a Mythos-class model, carried the same underlying concern. Mythos 5, a version of Fable 5 without some of the safeguards, was scheduled to be released to a more limited group of institutions for cybersecurity purposes.

The Broader Pattern

The administration's scrutiny has not stopped at Anthropic. Bloomberg also reported that OpenAI said it would limit access to a preview version of its GPT-5.6 model to a handful of select, government-approved partners, following similar pressure from the administration over the model's capabilities. The Anthropic situation appears to have established a template: the government flags a model, the company restricts access, and a negotiated clearance process follows.

Critics of the export-control approach argue that restricting AI models based on vague security concerns, without a transparent legal standard or due process, sets a precedent for government control over private technology development that could chill innovation and is effectively impossible for companies to challenge through normal channels. Anthropic, they note, had already voluntarily limited Mythos 5 distribution before Lutnick's letter arrived. In its own blog post announcing the disabling of the two systems, Anthropic warned that if the government's standard "was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers" and cautioned that "perfect jailbreak resistance is not currently possible for any model provider."

That concern is legitimate. The counterpoint, equally legitimate, is that a model capable of autonomously finding cybersecurity vulnerabilities at scale presents a national security risk that voluntary guardrails alone cannot adequately address, particularly when foreign actors could access the model through any one of thousands of global customers. The administration's position is that the risk is real enough to require government verification, not company self-attestation.

What Remains Unresolved

Neither side has said publicly what specific security commitments would satisfy the administration's concerns. The people familiar with the talks told Bloomberg that U.S. officials assured Anthropic the restrictions would be lifted once those concerns were addressed, but the precise criteria have not been disclosed. It also remains unclear what steps Anthropic would have to take to satisfy the government's concerns that it is possible to bypass guardrails on the Fable 5 model.

The open question is whether inter-agency alignment happens quickly or bogs down. Lutnick's Commerce Department appears ready to move. Whether the White House, the intelligence community, and other involved agencies reach the same conclusion on the same timeline is the variable that determines when, or whether, Fable 5 and Mythos 5 get back online for global users.